An advertising platform can only optimise towards outcomes it can observe. When half your buyers pick up the phone instead of filling in a form, the platform does not see a slower conversion. It sees no conversion at all, and it quietly moves your budget towards the people who behave in the way it can measure.
Answer first
If a meaningful share of your sales start with a phone call or an email, do three things in this order. Put a call tracking layer on the numbers people actually dial, using dynamic number insertion so the call carries the campaign that produced it. Make sure whoever answers records the outcome somewhere a machine can read. Then import only the qualified conversations back into the advertising platform as conversions, with a value. The first step makes calls visible, the second makes them meaningful, and the third is the one that changes where your money goes.
The bias is the real damage
The missing data is not the worst part. The worst part is what the platform does with its absence.
Automated bidding is a feedback loop. It buys more of whatever produced the outcomes you fed it, and it learns from every conversion you send and every one you do not. If form submissions are the only conversion in the account, the system gets better every week at finding people who like forms. Those are not the same people as the ones who ring you because the purchase is expensive, complicated or urgent.
So the account does not simply under-report. It systematically defunds the campaigns, keywords and audiences that produce your best buyers, and it does it while showing you an improving cost per lead. That is the uncomfortable version: the numbers can get prettier as the business gets worse.
In one luxury vacation rental account, 61 percent of bookings began with a phone or email conversation with staff. Every optimisation decision in that account had been made on the minority.
What tracking a call actually means
There are three levels, and most businesses stop one short of the one that matters.
Level one, a call happened. A tracking number on the website tells you volume. It is better than nothing and it changes no decisions, because you still cannot say which advertisement caused it.
Level two, where the call came from. Dynamic number insertion swaps the number shown on the page according to how the visitor arrived, so the call can be tied back to a campaign, a keyword, or a specific ad. This is the step that turns a phone into a measurable channel. Google’s own call reporting covers calls that begin from an ad. A third-party system also covers organic, referral and direct visitors, which in most service businesses is where the majority of the phone volume actually sits.
Level three, what the call was worth. This is where nearly everyone stops short. Knowing a campaign produced forty calls tells you nothing useful if eleven were existing customers, six were wrong numbers and two were suppliers. The account needs to know which conversations were real.
Grade the conversation before you feed it back
Sending raw call counts back to an advertising platform is worse than sending nothing, because it teaches the system to buy more calls of any quality. Cheap, bad calls are easy to find, and a bidding algorithm will find them enthusiastically.
Grade first. The crude version is duration, on the theory that a genuine enquiry rarely finishes in twenty seconds. The honest version is a human marking the outcome, because duration cannot tell a long complaint from a long sale. Whoever answers the phone needs one field to fill in, and it needs to take a moment rather than a minute, or it will not happen.
Then import only the qualified conversations, with a value where you know one. A booking worth four thousand and an enquiry that went nowhere should not be the same event.
The forgotten half: email and messaging
Everything above applies to the enquiries that arrive by email or WhatsApp, and those are usually less instrumented than the phone.
The mechanism is the same. Give the route a way to carry its source, whether that is a click identifier stored when the visitor lands, a link that records which page the message started from, or simply a shared inbox whose contents a machine can read. Then log the outcome the same way you would for a call.
The channel is never the point. The question is only ever whether the conversation is joined to the click that caused it and to the result it produced.
Where this genuinely cannot reach
Some conversations are unattributable, and pretending otherwise is how measurement work loses credibility.
Someone who saved your number two years ago, or dialled it from a directory listing, or was told about you by a friend, arrives with no digital trail at all. A tracking layer on your website cannot see a call that never touched your website. In the same rental account, 27 percent of bookings came through online travel agents, where no advertising click exists to find.
The right response is to state the floor rather than model over it. An account where you can prove three-quarters of the revenue and name the reason for the rest is in far better shape than one that claims everything and cannot show its working.
There is also a consent question worth settling before anyone presses record. Call recording law varies by jurisdiction and several require every party to agree. Tracking which campaign produced a call is a different and much lighter thing than recording its contents, and most businesses only need the first.
Where this sits in the chain
This is the second of the three places a chain of custody breaks between an advertisement and the money it produced, and the method we run repairs them in order. The first is the hand-off between domains, where the session restarts and the campaign is left behind. The third is the missing join between a click identifier and the record of the sale.
Repair them in that order. Fixing the hand-off is configuration. Seeing the conversation is a system and a habit. The habit is the harder half, because it depends on a person remembering to say what happened.